AISA-ArabicFC · Final Results

🥈

2nd place

Yalla

0.9273

Overall

🥇

1st place

Ṣaqr (صقر)

0.9279

Overall

🥉

3rd place

QMUL_AK

0.9253

Overall

Overall (Track A) = 0.40 · FnAcc + 0.60 · ArgEM

Rank	Team	Overall	FnAcc	ArgEM ★
🥇 1	Ṣaqr (صقر)	0.9279	0.9973	0.8816
🥈 2	Yalla	0.9273	0.9973	0.8806
🥉 3	QMUL_AK	0.9253	0.9938	0.8796
4	الشاهق	0.9189	0.9973	0.8666
5	Arabic Function Forge	0.9169	0.9893	0.8686
6	Thaka	0.9163	0.9893	0.8676
7	PuDyDG	0.9111	0.9973	0.8536
8	ArgTune Arabs	0.9089	0.9964	0.8506
9	Durham-Jazan	0.8928	0.9938	0.8255
10	وكلاء العربية	0.8870	0.9778	0.8265
11	MOSKA-LLM	0.8780	0.9689	0.8175
12	Saudi Legal FC	0.8729	0.9831	0.7994
13	farah	0.8690	0.8862	0.8576
14	GymMinds	0.8614	0.9964	0.7713
15	ToolCallers	0.8506	0.9636	0.7753
16	واجه	0.7904	0.9138	0.7081
17	Optimal	0.7232	0.9324	0.5838

🥈

2nd place

Yalla

0.9395

Overall

🥇

1st place

Ṣaqr (صقر)

0.9400

Overall

🥉

3rd place

QMUL_AK

0.9380

Overall

Overall (Track B) = 0.30 · FnAcc + 0.50 · ArgEM + 0.20 · ThinkRate

Rank	Team	Overall	FnAcc	ArgEM ★	ThinkRate
🥇 1	Ṣaqr (صقر)	0.9400	0.9973	0.8816	1.0000
🥈 2	Yalla	0.9395	0.9973	0.8806	1.0000
🥉 3	QMUL_AK	0.9380	0.9938	0.8796	1.0000
4	Arabic Function Forge	0.9311	0.9893	0.8686	1.0000
5	Thaka	0.9306	0.9893	0.8676	1.0000
6	PuDyDG	0.9260	0.9973	0.8536	1.0000
7	Sa74ll	0.9149	0.9947	0.8455	0.9689
8	Durham-Jazan	0.9109	0.9938	0.8255	1.0000
9	مَكين \| Makeen	0.9106	0.9911	0.8265	1.0000
10	Siraj	0.9101	0.9929	0.8245	1.0000
11	وكلاء العربية	0.9066	0.9778	0.8265	1.0000
12	Saudi Legal FC	0.8946	0.9831	0.7994	1.0000
13	farah	0.8936	0.8862	0.8576	0.9947
14	GymMinds	0.8718	0.9964	0.7713	0.9360
15	واجه	0.8282	0.9138	0.7081	1.0000
16	TIGER	0.8037	0.9929	0.6690	0.8569
17	Optimal	0.7137	0.9324	0.5838	0.7102

🥈

2nd place

الشاهق

0.9031

Macro-dialect

🥇

1st place

QMUL_AK

0.9075

Macro-dialect

🥉

3rd place

Ṣaqr (صقر)

0.8994

Macro-dialect

Track C = macro-average of per-dialect Overall (equal weight per dialect) · gap = max−min

Rank	Team	Macro	MSA	Gulf	Egy	Lev	Maghrebi	Gap
🥇 1	QMUL_AK	0.9075	0.935	0.915	0.932	0.897	0.859	0.077
🥈 2	الشاهق	0.9031	0.924	0.934	0.925	0.891	0.841	0.093
🥉 3	Ṣaqr (صقر)	0.8994	0.942	0.913	0.935	0.902	0.806	0.136
4	Thaka	0.8978	0.929	0.910	0.903	0.901	0.847	0.082
5	Yalla	0.8957	0.946	0.892	0.928	0.907	0.806	0.140
6	Arabic Function Forge	0.8847	0.933	0.890	0.933	0.880	0.788	0.145
7	PuDyDG	0.8800	0.930	0.902	0.904	0.876	0.788	0.142
8	ArgTune Arabs	0.8754	0.920	0.918	0.918	0.886	0.735	0.184
9	Sa74ll	0.8751	0.922	0.923	0.891	0.862	0.776	0.147
10	Siraj	0.8703	0.906	0.922	0.852	0.883	0.788	0.134
11	Durham-Jazan	0.8678	0.911	0.901	0.856	0.883	0.788	0.123
12	وكلاء العربية	0.8658	0.894	0.919	0.859	0.904	0.753	0.166
13	مَكين \| Makeen	0.8647	0.905	0.870	0.901	0.878	0.771	0.134
14	MOSKA-LLM	0.8592	0.888	0.877	0.869	0.874	0.788	0.100
15	Saudi Legal FC	0.8547	0.884	0.913	0.836	0.864	0.776	0.136
16	farah	0.8464	0.868	0.872	0.916	0.859	0.718	0.198
17	GymMinds	0.8335	0.876	0.865	0.828	0.881	0.718	0.163
18	ToolCallers	0.8247	0.868	0.841	0.821	0.858	0.735	0.132
19	TIGER	0.7804	0.810	0.807	0.779	0.788	0.718	0.092
20	واجه	0.7661	0.797	0.798	0.798	0.791	0.647	0.151
21	Optimal	0.6950	0.739	0.685	0.726	0.713	0.612	0.128

Metrics

FnAcc — function-name accuracy (all rows). ArgEM ★ — exact argument match (positive rows, Arabic normalization). ThinkRate — Track B reasoning-trace rate.
Track A = 0.40·FnAcc + 0.60·ArgEM · Track B = 0.30·FnAcc + 0.50·ArgEM + 0.20·ThinkRate.
Track C (dialect robustness) = the macro-average of each system's per-dialect Overall across MSA, Gulf, Egyptian, Levantine and Maghrebi — equal weight per dialect, so it rewards handling all dialects (not just the frequent ones). gap = max−min across dialects (lower = more consistent).
Fair Arabic normalization is applied identically to predictions and gold; the scorer is open in this Space. Top submissions were verified via model + code.

AISA-ArabicFC — Final Leaderboard

Metrics

Next